Our Dynamic World: Dense Land Cover Data & Stories of Shifting Landscapes
Deep dive into the Global Land Cover Fine Classification System dataset (GLC_FCS30D) with a 30m resolution, 35 classes, 26-time steps, and covering 1985-2022
While the GEE Community catalog primarily contains published datasets, some intriguing additions like the GLC_FCS30 have been included based on user requests and interest. This new preprint dataset offers a chance to explore long-term land cover changes at an unmatched resolution. The GLC_FCS30 captures global shifts across 35 classes every 5 years from 1985-2000, and annually thereafter - all at a granular 30-meter scale. The GLC_FCS30D dataset was created using continuous change detection applied to the full Landsat image archive on Google Earth Engine.
The top two panels showcase forest change in Brazil from 1985 to 2022, revealing extensive deforestation over nearly 40 years. The bottom panel documents urban expansion around Beijing over the same time period.
Stable areas were identified to derive training samples for locally adaptive classification models. These models were used to update land cover for changed pixels. A temporal consistency optimization refined the maps and suppressed false changes. The dataset was validated using over 84,000 global samples, achieving 73-81% accuracy. You can read the preprint here for further details.
Getting Hands-on with the Data 👩💻
The datasets can be readily accessed from the Zenodo repository and now from the Awesome GEE Community catalog. They include 961 total image tiles segmented into annual and 5-year products and respective earth engine image collections:
5-year maps (3 bands): Each band depicts land cover for 5 year periods (1985-1990, 1990-1995, 1995-2000)
Annual composites (22 bands): Each band represents land cover for a single year
Data Wrangling for Earth Engine Community Catalog ✨
The datasets come as zipped folders containing GeoTiffs. I kept the tiles as they were and created two separate collections for the 5-year and annual land cover data. ️ To make things even easier, I converted the class attribute RGB codes to hex codes for use in Earth Engine code. You can find the dataset page here
Both collections together equal about 700+ GB of total images in the collections with 961 tiles in each collection.
Example code from Awesome GEE Community Catalog here
Want to get involved and help shape the future of the GEE Community Catalog? Let's connect! Reach out on Linkedin and Github to share dataset ideas, provide feedback, and join the conversation.
Your support is invaluable for sustaining this project long-term. If you appreciate these efforts, please consider giving the GitHub repository a star ⭐️to increase the visibility of the work and spreading the word!
Stay tuned for this week's catalog updates highlighting some exciting new additions. 💡 Help out and become a sponsor of the community catalog and join me in building a vibrant community🌍
Appreciated:::::